Hangups in n_tty_read()

Hangups in n_tty_read()

[Johannes Weiner]

Hello Alan,

I regularly get hung terminals when ^Cing a foreground process.  The
terminal won't show any sign of input or output anymore then.

Here is an excerpt from sysrq+w:

Jul 16 17:33:10 chigurh kernel: [20907.892444] conf          S 0000000000000000     0 10946  10888 0x00000000
Jul 16 17:33:10 chigurh kernel: [20907.892444]  ffff88000e1efd58 0000000000000082 ffff88000e1efcc8 ffffffff8103423b
Jul 16 17:33:10 chigurh kernel: [20907.892444]  0000000000004000 0000000000012240 000000000000dc78 ffff880001639240
Jul 16 17:33:10 chigurh kernel: [20907.892444]  ffff88000e1efce8 ffff880024688000 ffff88002f8651c0 ffff880024688378
Jul 16 17:33:10 chigurh kernel: [20907.892444] Call Trace:
Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff8103423b>] ? try_to_wake_up+0x29e/0x2b0
Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff8103423b>] ? try_to_wake_up+0x29e/0x2b0
Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff8134f675>] schedule_timeout+0x26/0x190
Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff811ea909>] n_tty_read+0x496/0x74d
Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff8103424d>] ? default_wake_function+0x0/0xf
Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff811e65a5>] tty_read+0x7d/0xba
Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff8109a33b>] vfs_read+0xab/0x122
Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff8109a476>] sys_read+0x47/0x6f

Killing the hung foreground process from another terminal works.  But
it won't recover the screwed terminal and the terminal does not show
any reaction to it (like a prompt) even when the process is gone.

I remember one hang without X where the cursor-blinking was still
affected by keyboard input.

Unfortunately, I hadn't had the chance to test if input comes through
by blindly executing something disk heavy and watch the LEDs or
similar, so I can't say for sure if it's just output or if input is
affected as well.

First kernel I noticed this with was something past .30.  And it
certainly happens with Linus' latest.

Any idea?

	Hannes

Re: Hangups in n_tty_read()

[Johannes Weiner]

On Thu, Jul 16, 2009 at 08:48:46PM +0200, Johannes Weiner wrote:
> Hello Alan,

Sorry, misspelt your email address (no copy-paste without pty's ;-)

> I regularly get hung terminals when ^Cing a foreground process.  The
> terminal won't show any sign of input or output anymore then.
> 
> Here is an excerpt from sysrq+w:
> 
> Jul 16 17:33:10 chigurh kernel: [20907.892444] conf          S 0000000000000000     0 10946  10888 0x00000000
> Jul 16 17:33:10 chigurh kernel: [20907.892444]  ffff88000e1efd58 0000000000000082 ffff88000e1efcc8 ffffffff8103423b
> Jul 16 17:33:10 chigurh kernel: [20907.892444]  0000000000004000 0000000000012240 000000000000dc78 ffff880001639240
> Jul 16 17:33:10 chigurh kernel: [20907.892444]  ffff88000e1efce8 ffff880024688000 ffff88002f8651c0 ffff880024688378
> Jul 16 17:33:10 chigurh kernel: [20907.892444] Call Trace:
> Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff8103423b>] ? try_to_wake_up+0x29e/0x2b0
> Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff8103423b>] ? try_to_wake_up+0x29e/0x2b0
> Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff8134f675>] schedule_timeout+0x26/0x190
> Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff811ea909>] n_tty_read+0x496/0x74d
> Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff8103424d>] ? default_wake_function+0x0/0xf
> Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff811e65a5>] tty_read+0x7d/0xba
> Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff8109a33b>] vfs_read+0xab/0x122
> Jul 16 17:33:10 chigurh kernel: [20907.892444]  [<ffffffff8109a476>] sys_read+0x47/0x6f
> 
> Killing the hung foreground process from another terminal works.  But
> it won't recover the screwed terminal and the terminal does not show
> any reaction to it (like a prompt) even when the process is gone.
> 
> I remember one hang without X where the cursor-blinking was still
> affected by keyboard input.

Please note that this was in screen, too, so pseudo-terminals here as
well.

> Unfortunately, I hadn't had the chance to test if input comes through
> by blindly executing something disk heavy and watch the LEDs or
> similar, so I can't say for sure if it's just output or if input is
> affected as well.
> 
> First kernel I noticed this with was something past .30.  And it
> certainly happens with Linus' latest.
> 
> Any idea?
> 
> 	Hannes